Re: [PATCH] scsi: fix sense_slab/bio swapping livelock

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Mon, 2008-04-07 at 18:51 +0100, Hugh Dickins wrote:
> On Mon, 7 Apr 2008, Hugh Dickins wrote:
> > On Sun, 6 Apr 2008, James Bottomley wrote:
> > > 
> > > This was sort of accidentally fixed in scsi-misc by commit 
> > > 
> > > commit c5f73260b289cb974928eac05f2d84e58ddfc020
> > > Author: James Bottomley <James.Bottomley@HansenPartnership.com>
> > > Date:   Thu Mar 13 11:16:33 2008 -0500
> > > 
> > >     [SCSI] consolidate command allocation in a single place
> > 
> > Thanks, yes, that looks a good substitute to me.
> > 
> > > Could you check that:
> > > 
> > > master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6.git
> > > 
> > > and see if it alleviates the problem? ... if so, we can work out which
> > > pieces to backport.
> > 
> > Precisely that patch seems appropriate to 2.6.25-rc8-git, so I'm now
> > running the test with just that applied to 2.6.25-rc8 (plus cfq rcu
> > fix).  Not quite what you asked, but...
> > 
> > Strictly speaking, it'd take a couple of days to be reasonably sure
> > that the livelock is gone (it appeared to reproduce quicker once I
> > moved to -rc8 plus cfq rcu fix; but I'm not entirely convinced that
> > wasn't just coincidence).  But if nothing bad appears overnight,
> > let's assume your patch is the one to push: I'll report tomorrow.
> 
> Right, as expected, that seems to run fine: I've seen no problem with
> it in 17 hours - beyond, of course, all the page allocation failures
> which will plague such tests until something else is changed.
> 
> Somewhat irrelevant since Linus put my patch into 2.6.25-rc8-git
> (probably worth it for the various issues noted in the comment);
> but reassuring for 2.6.26.
> 
> I'll stop that test now and put it to work on something else.

Well ... just remember that to merge scsi-misc with what Linus has now
done, I'm effectively reversing your patch, so testing current scsi-misc
is very valuable if you want the problem not to recur in 2.6.26 ...

James